Available online at www.sciencedirect.

com

Speech Communication 50 (2008) 366–381

www.elsevier.com/locate/specom

The sound of sarcasm

Henry S. Cheang *, Marc D. Pell
McGill University, School of Communication Sciences and Disorders, 1266 Pine Avenue West, Montreal, Quebec, Canada H3G 1A8

Received 4 April 2007; received in revised form 17 November 2007; accepted 19 November 2007

Abstract

The present study was conducted to identify possible acoustic cues of sarcasm. Native English speakers produced a variety of simple
utterances to convey four different attitudes: sarcasm, humour, sincerity, and neutrality. Following validation by a separate naı̈ve group
of native English speakers, the recorded speech was subjected to acoustic analyses for the following features: mean fundamental fre-
quency (F0), F0 standard deviation, F0 range, mean amplitude, amplitude range, speech rate, harmonics-to-noise ratio (HNR, to probe
for voice quality changes), and one-third octave spectral values (to probe resonance changes). The results of analyses indicated that sar-
casm was reliably characterized by a number of prosodic cues, although one acoustic feature appeared particularly robust in sarcastic
utterances: overall reductions in mean F0 relative to all other target attitudes. Sarcasm was also reliably distinguished from sincerity by
overall reductions in HNR and in F0 standard deviation. In certain linguistic contexts, sarcasm could be differentiated from sincerity and
humour through changes in resonance and reductions in both speech rate and F0 range. Results also suggested a role of language used by
speakers in conveying sarcasm and sincerity. It was concluded that sarcasm in speech can be characterized by a specific pattern of pro-
sodic cues in addition to textual cues, and that these acoustic characteristics can be influenced by language used by the speaker.
Ó 2007 Elsevier B.V. All rights reserved.

Keywords: Verbal irony; Sarcasm; Prosody; Acoustic cues

1. Introduction Verbal irony can be defined as expressions in which the

Communication and speech comprehension are heavily
dependent on the use of implicit information (Sabbagh,
1999). Speakers convey implicit information to listeners
by manipulating language and prosody (i.e., intonation
and stress patterns), among other features, to express a
particular message. The rules that govern how speakers
produce language are well-documented (e.g., Grice,
1975). In contrast, how prosodic cues are used to express
affective and attitudinal states has been studied much less.
One context in which prosodic features appear to play a
significant role is the communication of verbal irony, of
which a key subtype is sarcastic irony, i.e., sarcasm (Kreuz
and Roberts, 1993).
*
Corresponding author. Tel.: +1 514 398 8496; fax: +1 514 398 8123.
E-mail addresses: henry.cheang@mail.mcgill.ca (H.S. Cheang), marc.
pell@mcgill.ca (M.D. Pell).
intended meaning of the words is different from or the
direct opposite of their usual sense; these expressions serve
numerous functions in communication (see Gibbs, 2000;
Haverkate, 1990, for a description of the forms and func-
tions of verbal irony). Sarcasm is verbal irony that
expresses negative and critical attitudes toward persons
or events (Kreuz and Glucksberg, 1989). However, it bears
noting that while researchers typically refer to or study
‘‘verbal irony” or ‘‘irony”, they are generally referring to
the negative attitude projected by ironic speakers. Hence,
in many instances, the terms ‘‘verbal irony” and ‘‘sarcasm”
have been conflated (Capelli et al., 1990). To be explicit, the
focus of the present study is sarcasm because of its impor-
tance in communication. For example, sarcastic comments
are quite pervasive in conversation, perhaps because listen-
ers tend to find these remarks less threatening and more
polite than overtly critical statements (Dews et al., 1995;
Gerrig and Goldvarg, 2000; Jorgensen, 1996; Kumon-
Nakamura et al., 1995). Viewed in another way, sarcastic

0167-6393/$ - see front matter Ó 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.specom.2007.11.003
 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381
comments can act to highlight and enhance the critical
message intended by speakers (Colston, 1997). Overall,
the degree to which sarcastic comments are considered
polite or critical appears to vary as a function of the surface
form of the message; sarcastic compliments are perceived
as more mocking and less polite than direct compliments,
whereas sarcastic insults are perceived as more mocking
and more polite than direct insults (Pexman and Olineck,
2002). While not the focus of this study, a number of the-
ories have been put forth to account for the contexts and
linguistic mechanics under which speakers express the neg-
ative subtype of verbal irony, i.e., sarcasm (e.g., Clark and
Gerrig, 1984; Grice, 1975; Sperber, 1984).
A point of general agreement for which there exists rel-
atively little cohesive data is that speakers can signal their
sarcastic intent to listeners using the prosodic content of
speech. One explanation for the inconsistent amount of
data is that different researchers have been studying distinct
subtypes of verbal irony that may present with different
prosodic patterns which may in turn lead to the conclusion
that no true acoustic patterns of irony exist. Speakers may
be using non-standard shifts in communicative contexts
(i.e., changes in acoustic parameters, semantics, discourse,
and facial expression, among others) to signal different sub-
types of verbal irony (including sarcasm) by highlighting a
deviation from an expected or typical message to listeners
(Attardo et al., 2003; Bryant, in press; Bryant and Fox
Tree, 2005; Kreuz and Roberts, 1995; Wilson and Whar-
ton, 2006). An alternative perspective is that, in conjunc-
tion with situational context and vocabulary choice,
specific acoustic cues known collectively as the ‘‘ironic tone
of voice” help listeners to know when sarcasm is intended
(Clark and Gerrig, 1984). However, given the different sub-
types of verbal irony that exist (Gibbs, 2000), it may be
more accurate to collectively refer to cues marking sarcastic
speech as a sarcastic tone of voice (even though researchers
typically do not make such a distinction). In principle, such
a pattern of acoustic cues is similar to the predictable
changes in acoustic cues that are associated with many
affective and attitudinal states (Banse and Scherer, 1996).
Note also that listeners can accurately recognize emotions
(e.g., joy, anger) and certain attitudes (e.g., confidence,
politeness) when listening to semantically-meaningless
‘‘pseudo-utterances” which communicate these meanings
strictly through prosodic cues (Dara et al., in press; Mon-
etta et al., in press; Pell, 2006). In an analogous manner,
it is possible that speakers use a relatively consistent set
of acoustic markers in conjunction with linguistic and con-
textual cues to signal sarcastic intent in speech.
In one study, Bryant and Fox Tree (2002) extracted sar-
castic and non-sarcastic utterances from live radio conver-
sations to determine whether listeners could discern
sarcasm under different experimental conditions. In the
absence of their original contexts, the texts of sarcastic
and non-sarcastic target utterances were not found to be
significantly non-biasing toward or away from a sarcastic
interpretation when presented in a written format (i.e., in
367
the absence of prosodic cues); however, when these utter-
ances were heard by a separate group of participants, utter-
ances extracted from sarcastic contexts were rated as
significantly more sarcastic than those from non-sarcastic
contexts. As well, additional participants could successfully
match the spoken utterances to transcribed versions of
their original contexts. One final separate group of partic-
ipants rated the originally-sarcastic utterances as more sar-
castic regardless of the presence of written versions of the
original sarcastic contexts (Bryant and Fox Tree, 2002).
These results show that the intent of sarcastic utterances
can be recognized independent of their contexts, although
one shortcoming cited by the authors is that the effects of
verbal semantics and acoustic features for understanding
sarcasm cannot be truly dissociated with these methods.
Semantic cues (i.e., words or phrases) often signal sarcasm
in conjunction with prosodic cues (Bolinger, 1989; Haiman,
1998) and there are some phrases or words which may be
so closely tied to the sarcastic context that these expres-
sions can independently signal sarcasm (i.e., these expres-
sions become ‘‘enantiosemantic”, Haiman, 1998, p. 39).
Thus, the role of prosody for communicating sarcasm can-
not be determined easily from this study.
In a follow-up study, Bryant and Fox Tree (2005) inves-
tigated how sarcasm is perceived from spontaneous utter-
ances taken from radio shows when the stimuli were
content-filtered (i.e., semantic cues were rendered unintelli-
gible). They found that filtered utterances that conveyed
‘‘dripping sarcasm” (i.e., utterances in which the sarcastic
intent was conveyed in a semantically and prosodically
unambiguous fashion, p. 260) were correctly rated as more
sarcastic by listeners than non-sarcastic utterances filtered
in the same manner (Bryant and Fox Tree, 2005). Comple-
mentary to these findings, Rockwell (2000b) showed that
naı̈ve listeners can discern sarcasm from posed, although
not spontaneous, utterances that were content-filtered
prior to presentation. Rockwell’s (2000b) analyses further
revealed that listeners perceive sarcastic utterances as hav-
ing lower pitch (i.e., the perceptual correlate of fundamen-
tal frequency (F0)), slower tempo (i.e., the perceptual
correlate of speech rate) and greater loudness (i.e., the per-
ceptual correlate of amplitude), although these conclusions
were derived from analyses of perceptual attributes and
hence do not supply direct evidence of acoustic changes
which might correspond with sarcastic speech.
In the developmental literature, there are also compel-
ling indications that contextual and prosodic cues for
appreciating sarcastic intent in speech can be dissociated.
Studies which have required children of different ages to
recognize sarcasm on the basis of context alone, intonation
alone, or combinations of the two sets of cues have shown
a developmental progression in the differential use of sar-
castic cues. In some reports, sensitivity to sarcastic situa-
tional contexts appears to develop first in young children
(Ackerman, 1983, 1986; Winner and Leekman, 1991),
whereas others claim that sensitivity to sarcastic intonation
is acquired first (Capelli et al., 1990; Laval and Bert-
368 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381
Erboul, 2005; Laval, 2004). Regardless of which milestone
is attained first, these findings imply a functional dissocia-
tion in the cues which contribute to the expression of sar-
casm in speech, underscoring the importance of prosody
in this context and allowing for the possibility that a
uniquely sarcastic prosody exists alongside contextual
markers of this attitude.
To date, relatively little work focusing on sarcasm has
been done that specifically examined the acoustic cues of
sarcasm. Among the acoustic features most frequently
cited in this literature are: heightened pitch/F0 variation;
heightened loudness; alterations in speech timing (e.g.,
reduced speech rate or increased number of pauses); varied
changes in voice quality; extra nasal resonance; and mono-
tonic or lowered pitch/F0 (Cutler, 1974, 1976; Fonagy,
1971; Haiman, 1998; Mueke, 1969, 1978; Rockwell,
2000a; Schaffer, 1982). More recent work has shown exem-
plars of sarcasm to have very complex F0 manipulations
across productions (Attardo et al., 2003) or to have high
F0 levels (Laval and Bert-Erboul, 2005), or a combination
of high F0 levels, high amplitude and a voice quality that
could be characterized as ‘‘tight” (Anolli et al., 2002).
Many of these findings highlight the importance of tempo-
ral and pitch (F0) measures, although note the differences
in whether sarcasm is associated with a higher (Anolli
et al., 2002; Attardo et al., 2003; Laval and Bert-Erboul,
2005) or lower (Rockwell, 2000a) pitch/F0 in the literature.
Overall, it is difficult to infer what prosodic cues may be
most critical for marking sarcasm, a situation that has per-
haps arisen from the different methods used and/or differ-
ences in the languages studied across studies.
Another relevant consideration is that acoustic differ-
ences may characterize different subtypes of verbal irony.
In addition to investigating the acoustic profile of sarcasm,
Anolli et al. (2002) found significant differences in acoustic
features between exemplars meant to express sarcasm and
exemplars meant to represent humourous irony (i.e., con-
veying a positive and playful mood). Recall too, that in
addition to conveying sarcasm and playful humour, verbal
irony performs other functions as well (Colston and
O’Brien, 2000a,b; Kreuz et al., 1991). Previous descriptions
of acoustic markers of sarcasm may be confused with
acoustic markers of other forms of verbal irony (should
such cues exist) or other attitudes. We need to expand
and clarify our knowledge regarding the cues that likely
mark sarcasm.
1.1. The present study
In light of the evidence that prosody is important for
understanding sarcasm in speech, the present study was
designed to provide a comprehensive acoustic description
of sarcastic utterances in English and to evaluate whether
a consistent pattern of acoustic cues differentiates sarcasm
from other ‘‘attitudes” as previously suggested (Clark and
Gerrig, 1984; Haiman, 1998; Rockwell, 2000a). To achieve
this, we recruited speakers who produced utterances to
simulate sarcasm and three other ‘‘attitudes” (humour, sin-
cerity, neutrality) and then undertook detailed acoustic
analyses of those utterances which were reliably judged to
communicate the intended attitude by a group of listeners.
In this way, we sought to specify prosodic differences that
might distinguish sarcasm from verbal irony intended to
sound humourous as well as from sincere/neutral exem-
plars of the same utterance. We studied posed tokens of
sarcasm to ensure greater experimental control over our
stimuli and to permit a further manipulation of the type
of utterance produced; our acoustic analyses were per-
formed on utterances which included or excluded enanti-
osemantic terms (key phrases argued to be associated
with sarcasm) to examine whether acoustic cues varied
systematically due to the presence of this linguistic
information.
To compare our findings with the existing literature, our
analyses included previously-studied measures of F0
(mean, range, and standard deviation), amplitude (mean
and standard deviation), and speech rate (e.g., Anolli
et al., 2002; Bryant and Fox Tree, 2005; Rockwell,
2000b, 2005). As well, we included a measure of both voice
quality and nasal resonance which have each been cited as
important for expressing sarcasm, among other attitudes
(e.g., Cutler, 1974; Haiman, 1998; Schaffer, 1982) but for
which there are little empirical data. With respect to voice
quality, the harmonics-to-noise ratio (HNR) was chosen to
quantify potential differences between sarcastic utterances
and other attitudes. The HNR can be defined as the aver-
aged periodic (i.e., harmonic) component of a sound signal
divided by the corresponding averaged noise component
(Yumoto et al., 1982); this measure has been found to be
a robust and reliable measure of voice quality changes in
studies of vocal pathology and normal aging and correlates
well with perceptual and other objective evaluations of
voice quality in the wider literature (de Krom, 1995; Esken-
azi et al., 1990; Pereira et al., 2002; Yumoto et al., 1982). In
terms of nasality (i.e., resonance), one-third octave spectral
analysis was employed as a means of gauging these differ-
ences and their relationship to expressions of sarcasm. This
form of analysis, which is performed by applying a one-
third octave filter to the important frequencies of a steady
state portion of a vowel and measuring the resulting values,
is known to correlate well with expert ratings of the pres-
ence and degree of clinical hypernasality regardless of etiol-
ogy, patient age (pediatric or adult) or gender, and across
languages (Kataoka et al., 1996, 2001a,b; Lee et al., 2003,
2004; Yoshida et al., 2000). These additional acoustic mea-
sures allow the present literature on sarcasm to be extended
in a meaningful way.
Based on the literature reviewed, it was expected that
exemplars of sarcasm would differ significantly from exem-
plars of humour and sincerity on measures of F0, ampli-
tude, speech rate, voice quality, and nasality. While
disparities in the literature prevent precise statements
regarding the directionality of these cues, it was expected
that sarcastic utterances would display a lower F0, greater
 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381 369

F0 variability, greater amplitude, and reduced speech rate
when compared to utterances expressing the other attitudes
(especially corresponding utterances which were intended
to be sincere). The effects of communicating sarcasm or
other attitudes on our measure of voice quality and nasality
could not be predicted with any certainty from the litera-
ture. For example, while HNR is very sensitive to the pres-
ence of hoarseness (Yumoto et al., 1982), breathiness (de
Krom, 1995), and general vocal changes brought about
by pathology (Eskenazi et al., 1990; Pereira et al., 2002);
or normal aging (Ferrand, 2002), this measure has not been
applied to the investigation of sarcasm. The effects of sar-
casm on our measure of nasality were similarly exploratory
in nature.
2. Method
2.1. Stimuli production
2.1.1. Encoders and materials
Six native English speakers or ‘‘encoders” (three male
and three female; mean age in years: 21.43, SD: 1.63; mean
years of education: 15.33, SD: 1.63) were recruited to pose
expressions of four distinct attitudes (sarcasm, humour,
sincerity, and neutrality). As described below, each encoder
was recorded producing a total of 96 utterances (24 repre-
senting each of the four attitudes). With the exception of
neutrality, each recorded exemplar was generated in
response to a biasing sentence to promote relatively natu-
ralistic productions of a given attitude (i.e., as if the enco-
der were engaging in a scripted dialogue; see Table 1 for
examples of all recording materials).
Target utterances were constructed to control the
semantic content, syntactic content, number of syllables,
and word frequency of the vocabulary (which was
restricted to high-frequency words). As the primary pur-
pose was to isolate possible acoustic cues of sarcasm, the
text of each item was devised in a way that it could be spo-
ken with each of the four attitudes without any changes. To
facilitate productions of sarcasm, humour, and sincerity,
sets of biasing sentences were created which were suitable
to each attitude; for example, biasing sentences designed
to promote the expression of sarcasm included very obvi-
ous harsh, unfairly critical, or insulting cues (e.g., ‘‘horrid
woman” from Table 1). For humour, biasing sentences
incorporated playful cues or explicit mention of a friendly
relation (e.g., ‘‘your friend Shelley” from Table 1). To
elicit sincerity, biasing sentences did not contain any infor-
mation that suggested positive or negative associations;
neither were there playful nor insulting cues. No biasing
sentences were constructed to assist encoders to express
neutrality.
A secondary objective was to evaluate the possible
impact of different utterance forms on how speakers mod-
ulate the acoustic cues to express sarcasm. Therefore,
within each group of tokens that expressed the target atti-
tudes, the utterances took three different forms: utterances
characterized by ‘‘keyphrases” cited by some authors as
more likely to convey sarcasm than other comparable
terms; sentences without keyphrases; and ‘‘combined sen-
tences” composed of the keyphrases and the sentences
combined in one utterance. Four sets of the three phrase
types were constructed in total. The text of these exemplars
ranged from 2 to 3 syllables for keyphrases and 7 to 8 syl-
lables for sentences (review Table 1). When producing a
given attitude, the same biasing sentence was used to elicit
productions for each of the three associated phrase types.
Hence, for every exemplar in the ‘‘healthy lady” set in
Table 1, the biasing sentence (for sincerity) was ‘‘She runs
10 miles a day”. To ensure the suitability of recording

Table 1
Biasing sentences and elicited sentences forming dialogues across four attitudes (sarcasm, humor, sincerity, and neutrality)
Phrase Text of elicited exemplar Attitude Biasing sentences associated with phrase type cluster
type
A: I suppose; it’s a respectful Sarcasm Don’t you just love how your stupid mother-in-law always smirks and snorts loudly
gesture. when you misspeak?
B: It’s a respectful gesture. Humor Not everyday that you see a priest give the finger, is it?
C: I suppose. Sincerity It was nice of your supervisor to send flowers.
A: Is that so; she is a healthy lady. Sarcasm That horrid woman smokes a pack a day.
B: She is a healthy lady. Humor Your friend Shelley can’t even do a single pushup.
C: Is that so? Sincerity She runs 10 miles everyday.
A: Oh boy; he is a superior chef. Sarcasm Our moronic boss gave us all food poisoning.
B: He is a superior chef. Humor Your brother singed his eyebrows while making toast.
C: Oh boy. Sincerity Butch just won another cooking contest; this is his twentieth win in the last 3 years.
A: Yeah, right; what a spectacular Sarcasm The arrogant front-runner finished dead last.
result.
B: What a spectacular result. Humor Fascinating how she lost the eating contest to someone half her size huh?
C: Yeah, right. Sincerity He broke three records in that race!
‘‘A” items denote combined sentences, ‘‘B” items denote single sentences, and ‘‘C” items denote keyphrases. Note that no biasing sentences were provided
to encoders to facilitate their production of neutrality.
370 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381

materials, all biasing sentence/target utterance pairs were microphone (sampling rate of 44.1 kHz, 16 bit, mono; no
presented in written format to four native English speakers precautions were taken against possible anti-aliasing).
who judged the naturalness of the sentence pairs in a
pilot study. The pilot participants were instructed to rate
2.1.3. Perceptual validation study
the pairs as being ‘‘natural”, ‘‘somewhat natural”, or
A perceptual validation study was undertaken to estab-
‘‘unnatural”. Both biasing sentences and target utterances
lish the reliability of the encoders’ productions of specific
were revised on the basis of these ratings and submitted
attitudes prior to acoustic analysis. Sixteen native English
to different English speaking participants. Refinement
speakers (eight males, eight females; mean age in years:
continued until there was at least 75% agreement across
22.47, SD: 2.41; mean years of education: 16.38, SD:
raters that all sentence pairs reflected communicative inter-
1.71) were recruited from the same population as the
actions that were perceived as natural for the intended
encoders to judge the intended meaning of the recorded
attitude.
utterances (hereafter these participants will be referred to
as ‘‘decoders”). Before testing, decoders were provided
2.1.2. Recording procedure
the same descriptions of the attitudes that had been given
Each target utterance was recorded twice (non-sequen-
to the encoders. During the task, decoders were presented
tially) to express each attitude per encoder. During the
the full 576 utterances recorded from encoders in 19 blocks
recording session, the encoder was presented a series of
of approximately 30 utterances per block. Following pre-
cards which contained the biasing sentence (except for neu-
sentation of each utterance, each decoder identified the
trality) and an associated utterance form (i.e., a keyphrase,
conveyed attitude from among four alternatives (sarcasm,
a sentence, or a combined sentence). Encoders had to
humour, sincerity, and neutrality) in a forced choice task.
silently read the biasing sentence and then produce the tar-
Utterances were retained as exemplars for acoustic analysis
get sentence to communicate the intended target attitude.
if the rating agreement was two times chance levels or
Recording sessions always began with the elicitation of
greater (i.e., 50% or greater agreement across all 16 decod-
neutral utterances; in this condition encoders were asked
ers that a particular exemplar reflects a specific attitude).
to simply read the target sentence in a voice that did not
For those stimuli categorized in a way that was not
convey any particular affect or attitude. Once the entire
intended by the encoder, the original classification was
set of neutral utterances was recorded, utterances convey-
dropped in favor of the decoders’ collective perceptual rat-
ing the remaining attitudes were recorded in a fixed ran-
ings and that exemplar was recoded. This strategy was pur-
dom sequence which was the same for all encoders. The
sued to help meet the primary goal of evaluating what
total number of exemplars recorded in the experiment
acoustic features signal sarcastic speech. There are possible
was 576 (i.e., 6 encoders  4 attitudes  4 items  3 utter-
mismatches between intended meaning and perceived
ance forms  2 repetitions).
meaning in exemplars of sarcastic speech (Rockwell,
At the onset of recording, encoders were provided with
2000a). Hence, it was important to ensure acoustic analyses
the definitions of each attitude and given brief, standard-
were carried out on utterances that were robustly recog-
ized descriptions of situations under which interlocutors
nized by a range of decoders as expressing a particular
are likely to express sarcasm, humour, or sincerity (e.g.,
attitude.
‘‘people use sarcastic utterances to respond to insulting
Subsequent to the perceptual validation study, we
comments directed at them”; ‘‘people use humourous state-
retained 489 utterances as ‘‘good” exemplars for acoustic
ments to be playful with friends”). Given the primary aim
analysis. As a whole, the utterances were representative
of the current study (i.e., evaluate the potential acoustic
of the target attitudes and were comparable across phrase
features of sarcastic verbal irony against other forms,
types. Results of the validation study are summarized in
including positive verbal irony), it was important to high-
Table 2.
light the fundamentally negative nature of sarcasm as the
negativity represents the chief distinction across the target
attitudes. The encoders were instructed to use the defini- 2.2. Acoustic analyses
tions, presented scenarios, and biasing sentences to guide
their enactments of the attitudes (see Appendix A for text All validated utterances were subjected to acoustic anal-
of all descriptive materials). To familiarize encoders with yses using Praat speech analysis software (Boersma and
the recording procedure, each completed several practice Weenink, 2007). As noted earlier, a number of acoustic
trials before commencing with the recordings proper. The parameters were selected for their potential importance
examiner did not provide any feedback or coach the enco- for differentiating sarcasm from other attitudes based on
der as to how a ‘‘proper” rendition of the required attitude previous research findings and trends; the measures
should be produced. Encoders were allowed to repeat their obtained from each exemplar were:
recordings to their satisfaction, and when this occurred, the
final utterance produced for each trial was considered for (a) Mean F0 (in Hz): computed for the utterance as a
analysis. All recordings were captured onto digital audio whole to index the relative effects of pitch height or
tape in a sound-attenuated booth using a high-quality register on different attitudes.
 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381 371

Table 2
Total number of utterances retained as exemplars for various acoustic analyses subsequent to validation study
Acoustic parameter Elicited phrase type Attitude Total per phrase type
Sarcasm Humor Sincerity Neutrality
F0, amplitude, speech rate, HNR Combined sentence 44 (70%) 23 (65%) 52 (68%) 42 (72%) 161
Single sentence 11 (67%) 22 (64%) 106 (72%) 39 (68%) 178
Keyphrase 56 (70%) 20 (68%) 41 (72%) 33 (65%) 150
Total per attitude 111 65 199 114
Total per phrase type entered into
separate analyses of variance
Sarcasm Humor Sincerity Neutrality
1/3 octave Spectral value Combined sentence 13 (68%) 22 (69%) 35 (68%) 33 (73%) 103
Single sentence 10 (67%) 15 (67%) 60 (69%) 23 (68%) 108
Note that a grand total of 489 utterances were retained as exemplars for acoustic analysis. An utterance was retained as an exemplar if the rating
agreement was 2 chance levels or greater (i.e., 50% or greater agreement across all 16 raters that a particular exemplar reflected a specific attitude). Mean
agreement percentages for exemplars are in parentheses next to the total number of utterances retained per factorial combination of PHRASE TYPE and
ATTITUDE. Following normalization to neutral exemplars, data from 368 exemplars were entered into statistical analyses for all measures of F0,
amplitude, speech rate, and HNR.

(b) Standard deviation of F0 (in Hz): computed around
the mean F0 for the whole utterance as an index of
overall F0 variation.
(c) F0 range (in Hz): computed by subtracting the mini-
mum from the maximum F0 value of each exemplar
as a further index of F0 variation.
(d) Mean amplitude (in dB): computed for the utterance
as a whole to index the intensity or loudness of utter-
ances spoken with different attitudes.
(e) Amplitude range (in dB): computed by subtracting
the minimum amplitude value from the maximum
amplitude value of each exemplar to assess the degree
of variation in speech intensity or loudness.
(f) Speech rate: computed as the number of syllables in
each item divided by the total utterance duration
(in milliseconds).
(g) Harmonics-to-noise ratio (HNR, in dB): voice qual-
ity measure computed from the 50-ms stable, central
portion of vowels which were segmented from the
stressed syllables of each utterance (most vowels in
unstressed syllables were shorter than 50 ms; Yumoto
et al., 1982). The HNR is the ratio of the averaged
periodic component of a sound signal to the corre-
sponding averaged noise component and is measured
in decibels (dB) (Yumoto et al., 1982). Evaluating
HNR serves as a probe for voice quality changes,
rather than as a full test of voice shifts, as we only
evaluate the HNR of stressed vowels and since the
calculation of HNR is a measure of the total noise
in the signal rather than a careful evaluation of the
contributions of different acoustic perturbations
(such as jitter or shimmer, Yumoto et al., 1982).
(h) One-third octave spectral values (in dB): measure of
nasality performed on isolated 50-ms stable portions
of stressed /i/ vowels by applying a one-third octave
filter to 16 frequency regions that range from 125 to
6300 Hz (i.e., encapsulating the spectral region
between the F0 past the third formant of /i/ vowels).
Hence, values (measured in dB) from 16 one-third
octave bands were taken to form a spectrum of val-
ues. The resulting spectrum was then normalized for
statistical analysis; the process of normalization con-
sisted of subtracting the value of each one-third
octave band from the value of the one-third octave
band containing the F0 of the vowel (Kataoka
et al., 1996; Yoshida et al., 2000). For male encoders,
F0 was located in the band centered between 160 and
200 Hz, whereas F0 for female encoders could be
found in the band centered between 200 and
250 Hz. Successive bands were renumbered sequen-
tially with the band containing the F0 relabeled as
‘‘0”. Elevated nasal resonance in /i/ vowels has been
characterized by amplitude increases between first
and second formants (i.e., the spectral region defined
by bands 6–8) and energy decreases following the sec-
ond formants (i.e., the spectral region defined by
bands 10–14) (Kataoka et al., 1996, 2001a; Lee
et al., 2003; Yoshida et al., 2000).
The restricted choice of vowels for one-third octave
analyses is justified as /i/ is the only vowel for which reli-
able indices of nasal resonance have been comprehensively
determined using this method of analysis (Kataoka et al.,
1996, 2001a; Lee et al., 2003; Yoshida et al., 2000). Other
vowels will likely present with different (one-third octave)
spectral profiles in the presence of nasality (Beddor, 1993;
Maeda, 1993). Also, nasal coupling (and hence, nasal reso-
nance) would be immediately detectable in /i/ over other
vowels, given its narrow lingual constriction (Yoshida
et al., 2000). As the vowel /i/ only occurred in the sentence
portion of exemplars, data considered for analysis was col-
lected from /i/ vowels extracted from the words ‘‘she”,
‘‘he”, and ‘‘superior (/i/ from /pir/)” of sentence and com-
bined exemplars. It would have been preferable to have
included language in keyphrases that contained /i/, but
the extensive refinement employed to construct reliable
and valid exemplars of target attitudes rendered this impos-
sible. Consequently, we analyzed 108 /i/ vowels extracted
372 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381
from sentence exemplars and 103 /i/ vowels segmented
from combined sentence exemplars (the different number
of /i/ vowels available for analysis was due to different
numbers of sentence and combined sentence exemplars
being retained subsequent to validation; see Table 2).
2.3. Statistical analyses
Prior to statistical analysis, acoustical measures were re-
examined by the first author and a trained research assis-
tant. Corrections were made to a small subset of exemplars
for which the autocorrelation method produced obvious
errors (e.g., when null values were returned by the Praat
software due to automated script errors, mean F0 was taken
using commands available in the graphical user interface
instead – this occurred for less than 1% of all exemplars).
With the exception of one-third octave spectral values
(which were subjected to a previously established normali-
zation procedure as part of calculation), measures obtained
for exemplars of sarcasm, humour, and sincerity were each
normalized in reference to the same measure obtained for
exemplars of neutrality to allow comparisons across speak-
ers and exemplar types. Normalization also corrected for
unavoidable differences in microphone distance and instru-
ment recording levels that would have varied somewhat
across testing sessions. All individual values of F0, ampli-
tude, and speech rate were normalized (per speaker) in ref-
erence to the set of neutral exemplars by dividing the
difference between the averaged value of neutral exemplars
for a given phrase type from an individual data point by the
averaged value of neutral exemplars for a given phrase type
(e.g. [mean F0(one keyphrase exemplar of humour) mean
F0(all keyphrase exemplars of neutrality)]/mean F0(all keyphrase exem-
plars of neutrality)). In the case of HNR, all HNR values taken
from the stressed vowels of a particular exemplar were first
averaged prior to normalization in reference to neutral
utterances.
The normalized acoustic data (except for one-third
octave spectral data) from the 368 remaining exemplars
of sarcasm, humour, and sincerity were then subjected to
a series of ANOVAs. Each acoustic measure was analyzed
for the influences of PHRASE TYPE (keyphrase, sentence,
‘‘combined sentence”) and ATTITUDE (sarcasm, humour,
sincerity) in a repeated-measures design. For one-third
octave spectral data, separate one-way ANOVAs involving
the factor of ATTITUDE (sarcasm, humour, sincerity,
neutrality) were conducted on each normalized octave
band following Kataoka et al. (1996). Further differentiat-
ing the ANOVAs of the one-third octave spectral data
from the statistical analyses of the other acoustic data is
the fact that vowels taken from sentence exemplars
(n = 108) were analyzed separately from combined sen-
tence exemplars (n = 103). Data from exemplars of neutral-
ity had to be included in the ANOVAs of one-third octave
spectral values as normalization of these values did not
eliminate neutrality as a condition. Significant effects and
interactions were explored with Tukey’s HSD method
(a = 0.05). Effects that were subsumed by higher-order
interactions are not described.
3. Results
Table 3 furnishes the mean (un-normalized) measures of
F0, amplitude, speech rate and HNR which characterized
each attitude.
3.1. Fundamental frequency: mean, standard deviation, range
The impact of the three attitudes on parameters of F0
(mean, standard deviation, and range) is illustrated in
Fig. 1. The ANOVA performed on values of mean F0 yielded
a significant main effect for ATTITUDE (F(2, 359) = 16.57,
p < 0.001). Post hoc analyses revealed that sarcasm exem-
plars were significantly lower in mean F0 than humour and
sincerity exemplars (see Fig. 1a).
The ANOVA on F0 standard deviation yielded main
effects for PHRASE TYPE (F(2, 359) = 12.86, p < 0.001)
and ATTITUDE (F(2, 359) = 17.51, p < 0.001). Elabora-
tion of PHRASE TYPE indicated that keyphrase exem-
plars exhibited a significantly greater F0 standard
deviation than both sentence exemplars and combined sen-
tence exemplars. The main effect of ATTITUDE was
explained by the fact that sincerity exemplars were pro-
duced with greater F0 standard deviations than humour
and sarcasm exemplars (see Fig. 1b).
Analysis of F0 range yielded main effects for PHRASE
TYPE (F(2, 359) = 12.71, p < 0.001) and ATTITUDE
(F(2, 359) = 6.35, p = 0.002), and an interaction of these
factors (F(4, 359) = 3.53, p = 0.008). Post hoc inspection
of the interaction showed that the three attitudes were only
differentiated by F0 range for keyphrases, where sarcasm
had a more restricted F0 range than sincerity. For sarcasm,
F0 range did not differ as a function of the three phrase
types, whereas for humour and sincerity, F0 range tended
to be greater for keyphrases than for sentences (and com-
bined sentences in the case of sincerity, see Fig. 1c).1
3.2. Amplitude: mean, range
The ANOVA on mean amplitude yielded a main effect
of PHRASE TYPE (F(2, 359) = 5.58, p = 0.004). Post hoc
tests showed that sentence exemplars were produced with
significantly lower amplitude than keyphrase exemplars
and combined sentence exemplars. There was no significant
1
To briefly estimate the extent to which some of our acoustic measures
may have been interdependent, a series of three ANCOVAs (analyses of
covariance) were conducted to verify the observed effects of PHRASE
TYPE and ATTITUDE from the main analysis when potentially related
measures were entered as a covariate. Mean F0 was reanalyzed with (1) F0
standard deviation and (2) F0 range as a covariate, and mean amplitude
was reanalyzed with amplitude range as a covariate. All significant effects
observed in the original analyses remained unchanged, suggesting that
these cues operated in a relatively independent manner for communicating
sarcasm.
 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381 373

Table 3
Means of non-normalized values of (a) mean fundamental frequency (F0), (b) F0 standard deviation (SD), (c) F0 range (i.e., f 0maximum f 0minimum),
(d) mean amplitude, (e) amplitude range (i.e., maximum amplitude minimum amplitude), (f) speech rate, and (g) harmonics-to-noise ratio (HNR)
Utterance form Attitude (a) Mean F0 (b) F0 SD (c) F0 range
Female Male Female Male Female Male
Keyphrase Humor 221.7 (49.9) 148.7 (19.6) 40.8 (33.3) 24.1 (6.0) 125.1 (90.3) 81.7 (21.6)
Neutrality 203.5 (26.0) 130.1 (10.7) 14.6 (7.1) 11.8 (5.1) 51.7 (23.3) 44.7 (15.3)
Sarcasm 193.8 (23.2) 136.4 (12.8) 21.2 (9.1) 20.3 (8.6) 76.2 (31.8) 69.3 (26.7)
Sincerity 224.4 (39.2) 162.6 (28.2) 41.8 (23.2) 30.9 (11.0) 138.0 (78.0) 95.5 (34.2)
Sentence Humor 209.2 (28.3) 167.5 (24.2) 30.8 (7.0) 24.6 (5.8) 112.5 (22.4) 93.2 (14.4)
Neutrality 195.7 (20.7) 134.8 (17.0) 19 (15.7) 12.8 (6.9) 85.3 (79.0) 58.8 (30.6)
Sarcasm 189.1 (15.1) 125.3 (13.1) 23.5 (7.4) 15.9 (5.1) 90.7 (22.6) 56.7 (19.0)
Sincerity 220.8 (34.2) 145.5 (13.2) 33.9 (19.9) 25.6 (10.3) 117.2 (63.8) 94.6 (36.0)
Combined sentence Humor 203.4 (12.9) 157.5 (21.2) 32.2 (13.1) 29.1 (7.9) 176.9 (123.2) 105.8 (25.8)
Neutrality 197.9 (23.8) 136.9 (14.4) 17.0 (9.7) 19.4 (14.9) 93.5 (85.2) 92.2 (72.2)
Sarcasm 193.0 (16.9) 137.5 (8.7) 28.2 (14.0) 26.3 (13.6) 145.2 (103.7) 129.9 (90.6)
Sincerity 229.8 (30.1) 149.9 (16.2) 43.4 (17.5) 28.7 (10.9) 175.1 (68.8) 113.2 (48.4)

Utterance form Attitude (d) Mean amplitude (e) Amplitude range (f) Mean speech rate (g) Mean HNR
Female Male Female Male Female Male Female Male
Keyphrase Humor 72.8 (2.6) 72.2 (1.7) 28.3 (7.9) 28.9 (5.6) 3.1 (0.6) 3.6 (0.6) 14.7 (6.1) 16.1 (5.1)
Neutrality 67.7 (3.9) 67.5 (3.5) 32.7 (4.7) 32.3 (5.2) 3.9 (0.6) 3.4 (0.4) 16.0 (3.7) 14.5 (4.3)
Sarcasm 70.0 (5.9) 69.8 (4.1) 31.7 (11.5) 31.2 (6.5) 2.8 (0.6) 2.6 (0.5) 16.2 (4.7) 13.7 (4.8)
Sincerity 73.3 (4.1) 71.1 (3.4) 27.1 (5.1) 30.5 (6.0) 4.0 (0.8) 3.9 (0.7) 14.9 (3.8) 15.6 (3.9)
Sentence Humor 71.1 (2.9) 72.1 (2.2) 35.2 (11.8) 35.7 (5.0) 5.1 (0.5) 6.0 (0.8) 14.4 (3.0) 17.0 (4.5)
Neutrality 68.4 (3.0) 67.6 (3.1) 33.7 (4.7) 35.4 (6.9) 5.3 (0.8) 5.0 (0.7) 15.2 (4.5) 14.5 (4.5)
Sarcasm 70.9 (2.1) 66.4 (3.3) 36.2 (8.0) 36.6 (6.3) 4.8 (0.8) 4.9 (0.7) 13.9 (2.2) 14.7 (2.0)
Sincerity 69.8 (3.2) 67.8 (2.5) 34.6 (5.8) 34.8 (5.8) 5.5 (0.7) 5.3 (0.6) 15.7 (4.3) 13.1 (4.8)
Combined sentence Humor 68.7 (3.6) 71.1 (1.5) 43.4 (4.1) 42.0 (4.5) 4.0 (0.8) 5.4 (0.4) 15.2 (4.1) 16.7 (1.9)
Neutrality 65.6 (4.1) 65.9 (2.8) 41.4 (5.2) 42.3 (7.3) 4.4 (0.5) 4.2 (0.6) 18.0 (3.0) 15.7 (3.8)
Sarcasm 66.9 (3.6) 67.0 (3.1) 42.2 (6.2) 43.6 (3.9) 4.1 (0.5) 4.2 (0.7) 14.2 (3.6) 13.3 (2.5)
Sincerity 69.3 (3.7) 65.1 (2.4) 40.3 (5.5) 43.7 (5.0) 4.8 (0.7) 4.6 (0.4) 16.9 (2.9) 15.3 (2.7)
Overall standard deviation values are in parentheses.

main or interactive effect involving the ATTITUDE factor
(both p’s > 0.14).
The ANOVA performed on amplitude range yielded a
main effect of PHRASE TYPE (F(2, 359) = 7.77,
p < 0.001). This effect could be explained by the observa-
tion that keyphrase exemplars were produced with a
reduced amplitude range relative to exemplars of the other
two phrase types. Again, there was no significant main
effect or interaction involving ATTITUDE (both p’s > .20).
3.3. Speech rate
The ANOVA on speech rate values yielded main effects
for PHRASE TYPE (F(2, 359) = 30.46, p < 0.001) and
ATTITUDE (F(2, 359) = 26.19, p < 0.001) and an interac-
tion of these factors (F(4, 359) = 16.57, p < 0.001). Tests
performed on the interaction revealed that when producing
keyphrase utterances only, sarcasm was expressed at a
slower rate than both humour and sincerity, and humour
was also produced more slowly than sincerity. When
expressing both sarcasm and humour, speakers produced
keyphrases significantly slower than combined sentences
representing the same attitude (and sentences for sarcasm),
although sincerity was expressed at a similar rate across
phrase types.
3.4. Harmonics-to-noise ratio (HNR)
The ANOVA on HNR values revealed main effects of
PHRASE TYPE (F(2, 359) = 4.16, p = 0.02) and ATTI-
TUDE (F(2, 359) = 3.09, p = 0.047). In the former case,
combined sentence exemplars were characterized by lower
averaged HNR values relative to the two other phrase
types. The main effect of ATTITUDE was explained by
the fact that sarcastic messages exhibited a significantly
lower averaged HNR value than sincerity.
3.5. One-third octave spectral values
One-third octave spectral data are presented in Fig. 2.
One-way ANOVAs performed on normalized amplitude
data from /i/ vowels in sentence exemplars revealed signif-
icant differences for one-third octave spectral band 2
(F(3, 104) = 3.06, p = 0.03), band 3 (F(3, 104) = 2.93,
p = 0.04), band 10 (F(3, 104) = 4.05, p = 0.009), band 11
(F(3, 104) = 3.65, p = 0.02), band 13 (F(3, 104) = 2.80,
p = 0.04), and band 14 (F(3, 104) = 3.61, p = 0.02). The sig-
nificant difference found for analysis of band 2 in the sen-
tence exemplars can be accounted for by the greater
amplitude of /i/ vowels in sarcasm tokens relative to /i/
vowels in sincerity tokens. Differences found at band 3
374 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381

a Normalized Mean F0 Across Attitudes

0.20

Normalized Mean F0 0.15

0.10

0.05

0.00

-0.05
Humor Sarcasm Sincerity
Attitude

b Normalized F0 Standard Deviation Across Attitudes

2.0
Normalized F0 SD

1.5

1.0

0.5

0.0
Humor Sarcasm Sincerity
Attitude

c Normalized F0 Range Across Attitudes

1.0
Normalized F0 Range

0.5

0.0
Humor Sarcasm Sincerity
Attitude

Fig. 1. Various speaker manipulations of fundamental frequency (F0) across exemplars of humor, sarcasm, and sincerity. (a) Normalized mean F0 values
across attitudes; (b) normalized F0 standard deviation values across attitudes; (c) normalized F0 range values across attitudes.

are explained by /i/ vowels from humour tokens having
greater amplitude than /i/ vowels in sincerity tokens.
Greater amplitude of /i/ vowels in sarcasm tokens relative
to sincerity and neutral tokens accounted for the significant
differences found in analyses of bands 10, 11, and 14.
Taken as a whole, sentence exemplars of sarcasm were
characterized by a resonance profile different from the
other attitudes.
For /i/ vowels extracted from combined sentence exem-
plars, statistical analysis of one-third octave spectral bands
yielded significant differences in normalized amplitude for
the majority of bands. Specifically, differences in attitude
tokens were found in band 1 (F(3, 99) = 3.57, p = 0.02),
band 2 (F(3, 99) = 3.83, p = 0.01), band 3 (F(3, 99) = 7.84,
p< 0.001), band 4 (F(3, 99) = 2.86, p = 0.04), band 5
(F(3, 99) = 3.69, p = 0.01), band 6 (F(3, 99) = 3.18,
p = 0.03), band 7 (F(3, 99) = 3.09, p = 0.03), band 11
(F(3, 99) = 5.28, p = 0.002), band 12 (F(3, 99) = 4.68,
p = 0.004), band 13 (F(3, 99) = 5.72, p = 0.001), and band
14 (F(3, 99) = 3.87, p = 0.01). A number of significant dif-
ferences can be attributed to the greater amplitude of /i/
vowels in sarcasm tokens relative to sincerity exemplars
(bands 1, 2, 4, 6, and 7). Analysis of band 3 showed two
patterns: /i/ vowels in sarcasm tokens were greater in
 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381
One-third Spectral Profile of Combined Sentence Exemplars of Attitudes
10
Normalized Amplitude (dB) 0
-5
-10
-15
-20
-25
-30
* * * *
-35
0 1 2 3 4 5
Fig. 2. Normalized one-third octave spectral values (in dB) for combined sentence exemplars grouped by attitude. Note the clear difference in one-third
octave spectral profile of sarcasm relative to exemplars of the other attitudes. * denote bands for which significant differences in normalized amplitudes
were found across attitude exemplars.
amplitude than for those in sincerity and neutrality tokens,
and /i/ in neutrality tokens were higher in amplitude than /
i/ in sincerity tokens. Similarly, in band 5, both sarcasm
and neutrality tokens were produced with greater ampli-
tude than /i/ vowels in sincerity tokens. Resonance differ-
ences most distinguished the attitudes in bands 11 and
13; sarcasm was greater in amplitude than the other three
attitudes. Finally, analysis of bands 12 and 14 showed that
/i/ vowels in sarcasm tokens were produced with greater
amplitude than those in sincerity and neutrality tokens or
sincerity and humour tokens, respectively. Overall, analysis
of combined sentence tokens showed a more pronounced
pattern of resonance differences across attitudes than sen-
tence exemplars, with the profile for sarcasm being the
most distinct. Fig. 2 summarizes acoustic and statistical
findings for one-third octave spectral analyses for the com-
bined sentence exemplars.2
2
By normalizing our measures in reference to neutrality exemplars, our
data do not permit insight into how neutral utterances compared to sarcasm,
humour, and sincerity for our prosodic measures. To illuminate this briefly,
all data were re-normalized in the following manner: all individual values of
F0, amplitude, and speech rate, and the HNR data points derived for
individual exemplars were standardized per speaker in reference to his or her
entire set of exemplars by dividing the difference between the averaged value
of all exemplars from an individual data point by the standard deviation of
all exemplars (e.g., [mean F0(one exemplar of humour) mean F0(all exemplars)]/F0
SD(all exemplars)). The ANOVA on each acoustic measure was then rerun with
four levels of ATTITUDE (including neutrality). The newly-computed
statistical results were consistent with the results of our main analyses and
will not be reported. Rather, significant attitude-neutrality differences are
summarized as follows: sarcasm was characterized by greater F0 SD, greater
F0 range, greater amplitude and slower speech rate than neutrality; humour
had greater mean F0, greater F0 SD, greater F0 range, greater amplitude,
and slower speech rate than neutrality; sincerity had greater mean F0,
greater F0 SD, greater F0 range, greater amplitude, and faster speech rate
than neutrality.
375
Humor
Neutrality
Sarcasm
Sincerity
* * * * * * *
6 7 8 9 10 11 12 13 14 15 16
1/3 Octave Spectral Bands
4. Discussion
The present study sought to shed light on the prosodic
markers of sarcasm and to dissociate these cues, if possible,
from the effects of semantic cues by analyzing utterances
which did or did not contain keyphrases associated with
sarcasm. Acoustic changes that varied as a function of atti-
tude regardless of phrase type, therefore, could be inter-
preted as prosodic cues which are most important to the
expression of sarcasm (in reference to other attitudes in
the present study). Based on this assumption, our results
showed that overall reductions in mean F0, decreases in
F0 variation (standard deviation), and changes in HNR
(i.e., voice quality) were used most consistently by English
speakers to differentiate sarcasm from other attitudes.
These more central cues often coincided with further
changes in resonance and reductions in speech rate which
occurred in many, but not all, linguistic contexts as elabo-
rated below.
4.1. The most prominent cues for signalling sarcasm
A reduction in mean F0 was the most consistently
observed prosodic correlate of sarcasm in the present
study; this cue distinguished sarcasm from both humour
and sincerity regardless of phrase type (see Table 3). This
result is unsurprising given that adopting a monotonic or
lower F0 or pitch in speech are often cited acoustic cues
of sarcasm (Attardo et al., 2003; Fonagy, 1971; Haiman,
1998; Rockwell, 2000a). A tendency to raise or lower voice
pitch is known to be instrumental for signaling other atti-
tudinal and affective states as well (Banse and Scherer,
1996; Bänziger and Scherer, 2005), emphasizing the signif-
376 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381
icance of this cue for marking sarcasm and perhaps other
interpersonal intentions during speech.
In addition to mean F0, sarcastic exemplars displayed
smaller F0 standard deviation than exemplars of sincerity
overall; moreover, F0 range was similarly reduced for
sarcastic relative to sincere utterances although only for
keyphrase exemplar types. Our findings for these (overlap-
ping) measures are supported in part by descriptions of sar-
casm as exhibiting periods of reduced F0 or pitch
movements (Attardo et al., 2003; Fonagy, 1971, 1976).
However, much of the literature suggests that sarcasm is
marked by heightened F0/pitch variability as one of its
major prosodic cues in contrast to our findings (Attardo
et al., 2003; Cutler, 1974, 1976; Haiman, 1998). Pending
further research, one can argue that changes in the extent
of F0 variation produced by speakers is a relatively consis-
tent feature of sarcastic speech, although the direction of
these changes is not always uniform and/or this cue may
interact more extensively with the nature of the language
content (Bryant, in press).
Speakers also manipulated voice quality to convey sar-
casm; specifically, we found that exemplars of sarcasm
were differentiated from exemplars of sincerity by greater
amounts of noise (as inferred from lower HNR values in
sarcasm exemplars). As analysis of HNR in sarcastic
speech is a novel approach, we need to draw from studies
that have looked at the importance of voice quality manip-
ulations on listener perceptions of affect or mood to inter-
pret our findings (Gobl and Ni Chaisaide, 2003; Ladd
et al., 1985; Banse and Scherer, 1996; Scherer, 1986;
Scherer et al., 1984). First, it has been noted that differences
in voice quality are mostly associated with how listeners
perceive a speaker’s mood and attitudes rather than their
emotions, with the exception of anger (Gobl and Ni
Chaisaide, 2003). In particular, ‘‘harsh” or ‘‘tense” voice
quality tends to signal anger or negative states (e.g.,
‘‘stressed”, ‘‘hostile”) to listeners (Gobl and Ni Chaisaide,
2003), and there is evidence that listeners classify utterances
with a harsh or tense voice quality as conveying negative
moods such as annoyance, irritation, and hostility (Ladd
et al., 1985). These negative attributions may stem from a
cluster of predictable physiological changes which occur
in response to negative stimuli, such as heightened tension
of the respiratory system and vocal apparatus as well as
decreased salivation which are likely to produce a tense
voice quality (Scherer, 1986).
As sarcasm is associated with critical (i.e., negative) atti-
tudes on the part of the speaker, it is therefore noteworthy
that shifts in voice quality in our sarcasm exemplars were
characterized by increased noise (i.e., decreases in HNR).
Not only does this finding confirm the importance of voice
quality changes for expressing attitudes, it implies that the
inherently negative nature of sarcasm may be encoded
and expressed through similar acoustic changes as other
negative messages, i.e., by increasing the level of noise in
the voice. One can speculate that listeners become attuned
to such shifts in voice quality as one of the major vocal
cues for identifying sarcastic intent (Fonagy, 1971; Hai-
man, 1998; Mueke, 1969). However, further perceptual
and acoustic analyses are necessary to confirm this
possibility.
4.2. Effects of secondary cues and phrase type
There were other acoustic markers of sarcasm which
tended to be less pervasive across linguistic contexts and
may be more dependent on phrase type. Principally, we
found that a reduced speech rate and different resonance
patterns were associated with sarcasm, although these cues
were prevalent for specific phrase types. In the case of
speech rate, short keyphrase utterances such as ‘‘is that
so” tended to be spoken more slowly than both ‘‘sentence”
and ‘‘combined” phrase types overall, but interestingly,
these keyphrases were produced significantly slower when
the speaker intended to be sarcastic than either humourous
or sincere. Similarly, exemplars of humour tended to be
much slower when speakers produced keyphrases as
opposed to combined sentences. These findings suggest
that listeners are sensitive to (reduced) speech rate as a
potential cue to the two different subtypes of verbal irony
studied in the current investigation, albeit seemingly in an
interactive fashion with phrase type: the smaller the utter-
ance form, the more listeners appeared to respond to
reductions in speech rate as an indicator of verbal irony
(regardless of the presumed hostility or friendliness of sar-
casm and humour, respectively). Speech rate was not a dif-
ferentiating factor in how listeners identified attitudes for
longer phrase types (e.g., combined sentences). Thus, our
results extend previous data which have characterized sar-
castic speech as having a reduced speech rate (Anolli et al.,
2002; Bryant, in press; Cutler, 1974; Haiman, 1998; Mueke,
1978; Rockwell, 2000b) but further clarify that speech rate
differences likely vary with language usage and may be
more critical for signaling irony in short utterance types.
With respect to resonance, the one-third octave spectral
data indicate that /i/ vowels extracted from both sentence
and combined sentence exemplars of sarcasm were pro-
duced with significantly greater amplitude at nearly all cru-
cial frequencies. Sarcasm differed most consistently from
sincerity in sentence exemplars and presented a markedly
different resonance profile from all other attitudes in com-
bined sentence exemplars. Recall that previous investiga-
tions have discerned increases in amplitude in spectral
regions near the first formant concurrently with decreases
in amplitude in spectral regions near the second formant
as a marker of heightened nasal emission (Kataoka et al.,
1996, 2001a; Lee et al., 2003; Yoshida et al., 2000). It is
important to note, however, the differing resonance pat-
terns observed here are not consistent with patterns that
denote heightened levels of nasality; the observed spectral
profiles did not correspond with previously-established
hypernasal profiles (i.e., energy increases in bands 6–8
and energy decreases in bands 10–14) (Kataoka et al.,
1996, 2001a; Lee et al., 2003; Yoshida et al., 2000).
 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381
Rather, we found amplitude increases in nearly every band
indicating that exemplars of sarcasm were produced with a
different, but not nasal, resonance (and only for sentence
and combined sentence phrase types).
One explanation for these results is that nasalization
may have occurred within sarcasm exemplars in areas that
were not analyzed (i.e., vowels other than /i/), but this
appears unlikely. Given how /i/ is articulated (i.e., narrow
lingual constriction), and the fact that only /i/ in stressed
syllables was examined, any nasal emissions should have
been immediately apparent in the present one-third octave
spectral analyses. Previous claims that nasal resonance was
associated with sarcastic expression were not based on
objective measures (Cutler, 1974, 1976; Haiman, 1998)
and many factors are known to impact on listeners’ detec-
tion of nasality in speech (see Baken and Orlikoff, 2000, for
an overview). It is therefore possible that previous asser-
tions that sarcasm is associated with increased nasality
stem from perceived differences in nasal resonance due to
the patterns observed here.
Another possibility is that orofacial expressions of dis-
gust, which are characterized by a facial sneer or palatal
drop, occur in tandem with sarcastic speech and lead to
these resonance changes. This idea suggests that speakers
mimic disgust as a critical or supplementary means of pro-
jecting sarcasm to listeners (Cutler, 1974; Fonagy, 1971;
Haiman, 1998), although comparisons between disgust
and sarcasm are problematic since the vocal and orofacial
gestures associated with disgust are known to be highly
variable in speech and difficult to recognize by listeners
(Pell et al., in review). A different way that facial expres-
sions could affect resonance relates to the finding that sar-
casm is signalled by exaggerated facial gestures,
particularly affecting the mouth, or that speakers adopt a
‘‘blank face” to cue pending sarcasm (i.e., a purposefully
inexpressive or motionless facial expression, Attardo
et al., 2003; Cutler, 1974; Fonagy, 1976; Mueke, 1969;
Rockwell, 2001, 2005). Observed changes in resonance
may have been associated with such changes in facial
expressions, especially mouth and lip movements (Tartter
and Braun, 1994) or tongue contractions (Fonagy, 1971).
Future refinements in methodology could include video
recording and analysis of speakers’ facial, lip, or tongue
movements when producing sarcasm and other attitudes
which could then be compared to spectral measures.
Finally, in contrast to some previous claims (Anolli
et al., 2002; Bryant and Fox Tree, 2005; Rockwell, 2000a,
2005), no differences in amplitude or amplitude variability
distinguished sarcasm from the other attitudes here. Since
data on this measure are generally restricted to perceptual
impressions of loudness (Rockwell, 2000a, 2005), further
acoustic data will be needed to reconcile our findings since
the relationship between objective acoustic features and
their perceptual correlates is not precise (Sobin and Alpert,
1999; Williams and Stevens, 1972). The possibility that
loudness as a prosodic feature of sarcasm varies across lan-
guages also cannot be discounted; our data on English
377
speakers may differ from those of Anolli et al. (2002)
who found that Italian speakers produced sarcasm with
increased loudness. A cross-linguistic comparison would
be useful in determining whether amplitude or other acous-
tic parameters associated with sarcasm vary significantly by
language, which we are now undertaking (Cheang and Pell,
in preparation).
4.3. On the ‘‘ironic tone of voice”
Researchers agree that prosody is instrumental for com-
municating sarcasm in speech, but there are different opin-
ions regarding the acoustic cues associated with sarcastic
intent and the manner in which linguistic context interacts
with prosody to project a sarcastic message. Overall, our
data show that sarcastic messages in English possess reli-
able acoustic features (characteristic changes in F0 and
voice quality) which are produced by speakers irrespective
of the linguistic context of these expressions. These findings
are most consistent with the idea of an ironic tone of voice
(Clark and Gerrig, 1984; Mueke, 1969), or more precisely,
a sarcastic tone of voice (i.e, the existence of defining pro-
sodic features which are used to communicate sarcasm in
speech).
At the same time, the interaction of prosody with lan-
guage and context is also emphasized by our data (Cutler,
1974; Haiman, 1998); here, speakers appeared to provide a
set of ‘‘supplementary” acoustic cues when producing
short, keyphrase utterances. One can speculate that speak-
ers provide an enriched prosodic signal in this (or other)
language contexts to ensure that shorter excerpts of speech
are unambiguously treated as sarcastic by the listener when
semantic features of an utterance are less indicative of this
intent. In this way, expressing sarcasm may be similar to
other attitudes in that certain prosodic cues may be salient
independent of semantic information while other prosodic
features work in conjunction with the language content
(Scherer et al., 1984). If one accepts that prosodic cues
operate both independently and interactively with other
cues, the existence of a specific sarcastic prosody can be
reconciled with claims that verbal irony is marked by a
variety of communicative behaviours (Attardo et al.,
2003; Bryant, in press; Bryant and Fox Tree, 2005).
Our data also furnish clues about the acoustic properties
of sincerity and humour. Prominent acoustic differences
were most apparent between sarcasm and sincerity rather
than humour; measures of mean F0, F0 standard deviation,
and HNR clearly distinguished these two attitudes regard-
less of linguistic context (sincere utterances displayed a
higher mean F0, greater F0 standard deviation, and less
noise in the signal). This result should come as little sur-
prise given the opposing speaker intentions underlying
the articulation of sarcasm (indirect, semantically-inconsis-
tent criticism) and sincerity (direct literal appraisal) which
are shown here to be distinguished by multiple acoustic
cues in speech. In the case of humour, we found no evi-
dence of a highly contrastive prosodic pattern from the
378 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381
other attitudes, in line with claims that humourous intent is
normally communicated through a combination of factors
or cues. For example, some explicit play cue must alert the
listener to preclude seriousness (Berlyne, 1972; Suls, 1983),
and once a play context is established, the meaning of
humourous utterances can be determined when expecta-
tions generated by the speaker are violated in a variety of
ways (Suls, 1983, 1972; Brownell et al., 1983). In light of
these factors, it may have been difficult to capture how
prosody is used to convey humour with our approach,
and it is also likely that speakers had greater difficulty to
convey humourous intent in single, posed utterances as
suggested by the relatively low number of perceptually-
valid humour exemplars retained from the validation study
(review Table 2). Nonetheless, exemplars of sarcasm and
humour were found to be distinct on certain acoustic mea-
sures and it would be prudent of future researchers to eval-
uate different subtypes of verbal irony separately rather
than as a uniform class (Anolli et al., 2002; Bryant, in
press).
Finally, our inclusion of ‘‘enantiosemantic” phrases in
the text of our tokens merits some commentary. Based
on results of our validation study, it is readily apparent
that there was an uneven distribution in the number of
exemplars perceived as conveying each attitude according
to phrase type; the presence of keyphrases tended to facil-
itate perception of sarcasm while their absence facilitated
perception of sincerity (review Table 2). This pattern of
findings is consistent with previous (empirically untested)
claims that certain phrases become associated with sarcas-
tic speaker intent (Haiman, 1998), again underscoring the
significant role of lexical-semantic information in the
expression of sarcasm. Nonetheless, it bears re-emphasiz-
ing that attitudes other than sarcasm could be conveyed
and recognized from each keyphrase in our validation
study; also, for those utterances which were retained,
speakers executed changes in F0 and voice quality to mark
sarcasm irrespective of whether particular keyphrases were
present or absent, highlighting the potential independence
of these prosodic features as a central property of sarcastic
speech.
4.4. Limitations and future directions
The present study analyzed non-spontaneous (‘‘posed”)
speech, and while great care was taken to ensure that ade-
quate exemplars of each attitude were subjected to analysis,
the literature would benefit from future studies which
investigate the production and comprehension of sarcasm
in both controlled and spontaneous situations (cf. Bryant
and Fox Tree, 2005; Rockwell, 2000b). Nonetheless, we
do not believe that differences in stimulus acquisition
would affect our main conclusions about the acoustic form
of sarcasm since previous perception studies have shown
that listeners can recognize posed or spontaneous sarcasm
(Bryant and Fox Tree, 2005; Rockwell, 2000b). More
likely, our data represent instances in which speakers suc-
cessfully simulated coded prosodic features for expressing
sarcasm which resemble those used in spontaneous con-
texts and which were decoded by listeners in our validation
study.
One way to expand current knowledge regarding the
role of prosody for communicating sarcasm and other atti-
tudes is to study the effects of brain damage on these cru-
cial abilities. Patients with right hemisphere damage often
display impairments in producing and comprehending
prosody in affective and attitudinal contexts (e.g. Heilman
et al., 1984; Tompkins, 1991). Recently, we have shown
comparable difficulties to communicate emotions and ‘pro-
sodic attitudes’ such as confidence or politeness in patients
with Parkinson’s disease (Cheang and Pell, 2007; Pell and
Leonard, 2003; Pell et al., 2006). Deriving a better under-
standing of fundamental acoustic markers of attitudes in
speech and their biasing linguistic and contextual factors
would be germane to improve diagnosis and treatment of
many ‘‘pragmatic language disorders” in these patients.
To conclude, the present analyses strongly suggest that
there is a distinct pattern of acoustic cues associated with
sarcasm in speech, one that diverges most clearly from
expressions of sincerity. One cluster of cues, reduced F0
and HNR and decreased F0 standard deviation, robustly
marked sarcastic utterances, whereas changes in resonance
and reductions in speech rate appeared to be supplemen-
tary cues for marking sarcasm in certain linguistic con-
trasts. While this pattern of acoustic cues does not
precisely match those observed for other attitudes or affec-
tive states (Banse and Scherer, 1996; Mullenix, 2005), more
research is needed to clarify the extent to which the ‘‘sound
of sarcasm” is uniquely specified in speech.
Acknowledgements
This research was conducted as part of the first author’s
doctoral research and was supported in part by a CIHR-
K.M. Hunter Doctoral Training Award. Further funding
was provided by a Discovery Grant from the Natural Sci-
ences and Engineering Research Council of Canada to the
second author. The authors would also like to thank Paul
Boersma, Valter Ciocca, and Alice Lee for help in devising
PRAAT scripts to conduct one-third octave analyses, and
Marie Desmarteau and Andrea MacLeod for their assis-
tance in the preparation of the manuscript.
Appendix A
A.1. Summary of attitude descriptions given to all
participants
A.1.1. Sarcasm
Definition: A sharp, bitter, or cutting expression or
remark; a bitter gibe or taunt.
Description: There are two particular circumstances that
we would like to draw your attention to with respect to sar-
castic utterances. We want you to note that sarcasm is often
 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381
used by people as a retort to an insult. Sarcasm is also used
by people to criticize situations or other people who they find
unpleasant. The common theme that runs in these two situ-
ations is that there should be an element of malice. People
may use sarcasm under other conditions, but the situations
just described are what we would like you to keep in mind.
A.1.2. Humor
Definition: Something (action, speech, or writing) that is
ludicrous or absurdly incongruous or designed to be com-
ical or amusing.
Description: We would like you to keep two situations in
mind regarding humorous, playful utterances. One circum-
stance is that people attempt to make humorous comments
about unexpected but amusing or unthreatening events.
Humorous banter might also occur when people who have
friendly relations find themselves in happy and non-serious
interactions. The key element with such utterances is that
they are made under happy or potentially silly situations.
There are other circumstances under which people may
make humorous utterances, but these are the ones that
we would like you to keep in mind.
A.1.3. Sincerity
Definition: Free from pretence or deceit; genuine, hon-
est, frank.
Description: With respect to sincere utterances, we are
referring to situations where people are expressing what
they are honestly feeling or thinking on a subject for which
they have no personal connection or stake. That is, speak-
ers mean exactly what they are saying; they are not trying
to project any hidden messages or intentions. Sincere utter-
ances are different from neutral utterances.
A.1.4. Neutrality
Definition: Displaying or containing no overt emotion;
dispassionate, detached.
Description: Neutral utterances do not convey any emo-
tion or mood. That is, neither attitude nor intention is
expressed. Sincere utterances and neutral utterances are
two wholly different things.
A.2. Instructions specific to encoders
‘‘We are trying to understand how people communicate
their moods verbally. We would like you to act out a
number of sentences with certain moods so that we
may record and analyze them.
You will be shown pairs of sentences. Please read both
sentences silently first. Imagine that the first sentence
is directed at you from another speaker and that the sec-
ond sentence in the pair is your response. I want you to
act out the second sentence. Importantly, I want you to
say the sentence in a way that you feel is as natural
sounding as possible; do not be theatrical. Essentially,
say it the way you would if you were in that situation.
379
There are very specific moods that we would like you to
enact depending on the sentence. We will tell you which
mood we would like you to enact before we present you
with the sentences. The first sentence in the sentence pair
will always be designed to help evoke the mood that we
would like you to act out.
[Then the participant was provided with a description of
an attitude prior to the recording of that particular atti-
tude; see above].”
A.3. Instructions specific to decoders
‘‘You will be hearing sets of about 30 sentences in a row.
Each set of sentences will be followed by a short break.
For every sentence, the speaker is trying to express
either a sarcastic, humorous, or sincere attitude. Alter-
nately, the speaker is attempting to express no attitude
at all, that is, the sentence sounds neutral. Please note
that you will be hearing many different voices and that
not all speakers will sound the same.
[Then the participant was provided with descriptions of
all attitudes; see above].
While you listen to each sentence, you will also see the
names of these attitudes on the computer screen. Your
task is to judge which attitude the speaker is trying to
express. It is important that you try to listen to how each
speaker sounds rather than paying specific attention to
the words being spoken. After you have heard the whole
sentence, choose the attitude that is closest to what you
hear by clicking the label on the computer screen with
your mouse. Please do not overly rush an answer; the
computer will not move on to the next sentence if you
try to answer before a sentence has finished playing.
We will begin with a short practice session. Ready?”
References
Ackerman, B., 1983. Form and function in children’s understanding of
ironic utterances. Journal of Experimental Child Psychology 35, 487–
508.
Ackerman, B., 1986. Children’s sensitivity to comprehension failure in
interpreting a nonliteral use of an utterance. Child Development 57,
485–497.
Anolli, L., Ciceri, R., Infantino, M.G., 2002. From ‘‘blame by praise” to
‘‘praise by blame”: analysis of vocal patterns in ironic communication.
International Journal of Psychology 37 (5), 266–276.
Attardo, S., Eisterhold, J., Hay, J., Poggi, I., 2003. Multimodal markers of
irony and sarcasm. Humor: International Journal of Humor Research
16 (2), 243–260.
Baken, R.J., Orlikoff, R.F., 2000. Clinical Measurement of Speech &
Voice, second ed. Singular Publishing Group Inc., New York.
Banse, R., Scherer, K.R., 1996. Acoustic profiles in vocal emotion
expression. Journal of Personality and Social Psychology 70, 614–636.
Bänziger, T., Scherer, K.R., 2005. The role of intonation in emotional
expressions. Speech Communication 46, 252–267.
Beddor, P.S., 1993. The perception of nasal vowels. In: Huffman, M.,
Krakow, R. (Eds.), Nasals, Nasalization, and the Velum. Academic
Press, San Diego, pp. 171–196.
380 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381
Berlyne, D.E., 1972. Humor and its kin. In: McGhee, P.E., Goldstein, J.H.
(Eds.), The Psychology of Humor. Springer-Verlag, New York, pp.
43–60.
Boersma, P., Weenink, D., 2007. Praat: doing phonetics by computer
(Version 5.0.02) [Computer program]. Retrieved 27 December 2007.
Available from: <http://www.praat.org/>.
Bolinger, D., 1989. Intonation and Its Uses: Melody in Grammar and
Discourse. Stanford University Press, Stanford, CA.
Brownell, H.H., Michel, D., Powelson, J., Gardner, H., 1983. Surprise but
not coherence: sensitivity to verbal humor in right-hemisphere
patients. Brain and Language 18, 20–27.
Bryant, G.A., Fox Tree, J.E., 2002. Recognizing verbal irony in
spontaneous speech. Metaphor and Symbolic Activity 17 (2), 99–117.
Bryant, G.A., Fox Tree, J.E., 2005. Is there an ironic tone of voice?
Language and Speech 48 (3), 257–277.
Bryant, G.A., in press. Prosodic contrasts in ironic speech. Discourse
Processes.
Capelli, C.A., Nakagawa, N., Madden, C.M., 1990. How children
understand sarcasm: the role of context and intonation. Child
Development 61, 1824–1841.
Cheang, H.S., Pell, M.D., 2007. An acoustic investigation of Parkinsonian
speech in linguistic and emotional contexts. Journal of Neurolinguis-
tics 20 (3), 221–241.
Cheang, H.S., Pell, M.D., in preparation. The acoustics of sarcasm in
Cantonese and English.
Clark, H.H., Gerrig, R.J., 1984. On the pretense theory of irony. Journal
of Experimental Psychology: General 113 (1), 121–126.
Colston, H.L., 1997. I’ve never heard anything like it: overstatement,
understatement, and irony. Metaphor and Symbol 12 (1), 43–58.
Colston, H.L., O’Brien, J., 2000a. Contrast and pragmatics in figurative
language: anything understatement can do, irony can do better.
Journal of Pragmatics 32, 1557–1583.
Colston, H.L., O’Brien, J., 2000b. Contrast of kind versus contrast of
magnitude: the pragmatic accomplishments of irony and hyperbole.
Discourse Processes 30 (2), 179–199.
Cutler, A., 1974. On saying what you mean without meaning what you
say. In: LaGaly, M.W., Fox, R.A., Bruck, A. (Eds.). Chicago
Linguistic Society, Chicago, IL, pp. 117–127.
Cutler, A., 1976. Beyond parsing and lexical look-up: an enriched
description of auditory sentence comprehension. In: Wales, R.J.,
Walker, E. (Eds.). Dik, S.C., Kooij, J.G. (Eds.), North-Holland
Linguistic Series, Vol. 30. North-Holland Publishing Company,
Amsterdam, pp. 133–149.
Dara, C., Monetta, L., Pell, M.D., in press. Vocal emotion processing in
Parkinson’s disease: reduced sensitivity to negative emotions.
Neuropsychology.
de Krom, G., 1995. Some spectral correlates of pathological breathy and
rough voice quality for different types of vowel fragments. Journal of
Speech and Hearing Research 38 (4), 794–811.
Dews, S., Kaplan, J., Winner, E., 1995. Why not say it directly? The social
functions of irony. Discourse Processes 19 (3), 347–367.
Eskenazi, L., Childers, D.G., Hicks, D.M., 1990. Acoustic correlates of
vocal quality. Journal of Speech and Hearing Research 33 (2), 298–
306.
Ferrand, C.T., 2002. Harmonics-to-noise ratio: an index of vocal aging.
Journal of Voice 16 (4), 480–487.
Fonagy, I., 1971. Synthèse de l’ironie. Phonetica 23, 42–51.
Fonagy, I., 1976. Spoken language: dynamic and changing. Journal de
Psychologie Normale et Pathologique 73 (3–4), 273–304.
Gerrig, R.J., Goldvarg, Y., 2000. Additive effects in the perception of
sarcasm: situational disparity and echoic mention. Metaphor and
Symbol 15 (4), 197–208.
Gibbs Jr., R.W., 2000. Irony in talk among friends. Metaphor and Symbol
15 (1–2), 5–27.
Gobl, C., Ni Chaisaide, A., 2003. The role of voice quality in commu-
nicating emotion, mood and attitude. Speech Communication 40, 189–
212.
Grice, H.P., 1975. Logic and conversation. In: Cole, P., Morgan, J.L.
(Eds.), Speech Acts, Vol. (3). Syntax and Semantics, pp. 41–58.
Haiman, J., 1998. Talk is Cheap: Sarcasm, Alienation, and the Evolution
of Language. Oxford University Press, Oxford.
Haverkate, H., 1990. A speech act analysis of irony. Journal of Pragmatics
14, 77–109.
Heilman, K.M., Bowers, D., Speedie, L., Branch Coslett, H., 1984.
Comprehension of affective and nonaffective prosody. Neurology 34,
917.
Jorgensen, J., 1996. The functions of sarcastic irony in speech. Journal of
Pragmatics 26, 613–634.
Kataoka, R., Michi, K., Okabe, K., Miura, T., Yoshida, H., 1996.
Spectral properties and quantitative evaluation of hypernasality in
vowels. Cleft Palate: Craniofacial Journal 33, 43–50.
Kataoka, R., Warren, D.W., Zajac, D.J., Mayo, R., Lutz, R.W., 2001a.
The relationship between spectral characteristics and perceived hyper-
nasality in children. Journal of the Acoustical Society of America 109,
2181–2189.
Kataoka, R., Zajac, D.J., Mayo, R., Lutz, R.W., Warren, D.W., 2001b. The
influence of acoustic and perceptual factors on perceived hypernasality
in the vowel. Folia Phoniatrica et Logopaedica 53, 198–212.
Kreuz, R.J., Glucksberg, S., 1989. How to be sarcastic: the echoic
reminder theory of verbal irony. Journal of Experimental Psychology:
General 118 (4), 374–386.
Kreuz, R.J., Roberts, R.M., 1993. On satire and prosody: the importance
of being ironic. Metaphor and Symbolic Activity 8 (2), 97–109.
Kreuz, R.J., Roberts, R.M., 1995. Two cues for verbal irony: hyperbole
and the ironic tone of voice. Metaphor and Symbolic Activity 10 (1),
21–31.
Kreuz, R.J., Long, D.L., Church, M.B., 1991. On being ironic: pragmatic
and mnemonic implications. Metaphor and Symbolic Activity 6, 149–
162.
Kumon-Nakamura, S., Glucksberg, S., Brown, M., 1995. How about
another piece of pie: the allusional pretense theory of discourse irony.
Journal of Experimental Psychology: General 124 (1), 3–21.
Ladd, D.R., Silverman, K.E., Tolkmitt, F.J., Bergmann, G., Scherer,
K.R., 1985. Evidence for the independent function of intonation
contour type, voice quality, and F0 range in signaling speaker affect.
Journal of the Acoustical Society of America 78 (2), 435–444.
Laval, V., 2004. Pragmatique et langage non littéral: la compréhension des
demandes sarcastiques par les enfants. Psychologie Francßaise 49, 177–
192.
Laval, V., Bert-Erboul, A., 2005. French-speaking children’s understand-
ing of sarcasm: the role of intonation and context. Journal of Speech,
Language, and Hearing Research 48, 610–620.
Lee, A.S.-Y., Ciocca, V., Whitehill, T.L., 2003. Acoustic correlates of
hypernasality. Clinical Linguistics and Phonetics 17 (4–5), 259–264.
Lee, S.Y.A., Ciocca, V., Whitehill, T., 2004. Spectral analysis of
hypernasality. Journal of Medical Speech-Language Pathology 12,
173–177.
Maeda, S., 1993. Acoustics of vowel nasalization and articulatory shifts in
French nasal vowels. In: Huffman, M., Krakow, R. (Eds.), Nasals,
Nasalization, and the Velum. Academic Press, San Diego, pp. 147–167.
Monetta, L., Cheang, H.S., Pell, M.D., in press. Understanding speaker
attitudes from prosody by adults with Parkinson’s disease. Journal of
Neuropsychology.
Mueke, D.C., 1969. The Compass of Irony. Methuen, London.
Mueke, D.C., 1978. Irony markers. Poetics 7, 363–375.
Mullenix, J., 2005. The perceptual processing of vocal emotion. In: Clark,
A. (Ed.), Psychology of Moods. Nova Science Publishers, Hauppage,
NY, pp. 123–139.
Pell, M.D., 2006. Cerebral mechanisms for understanding emotional
prosody in speech. Brain and Language 96 (2), 221–234.
Pell, M.D., Leonard, C.L., 2003. Processing emotional tone from speech
in Parkinson’s disease: a role for the basal ganglia. Cognitive, Affective
and Behavioral Neuroscience 3, 275–288.
 H.S. Cheang, M.D. Pell / Speech Communication 50 (2008) 366–381
Pell, M.D., Cheang, H.S., Leonard, C.L., 2006. The impact of Parkinson’s
disease on vocal-prosodic communication from the perspective of
listeners. Brain Language 97, 123–134.
Pell, M.D., Alasseri, A., Paulmann, S., Kotz, S.A., in review. Factors in
the recognition of vocally expressed emotions: a comparison of three
languages.
Pereira, J.G., Cervantes, O., Abrahao, M., Parente Settanni, F.A.,
Carrara de, A.E., 2002. Noise-to-harmonics ratio as an acoustic
measure of voice disorders in boys. Journal of Voice 16, 28–31.
Pexman, P.M., Olineck, K.M., 2002. Understanding irony: how do
stereotypes cue speaker intent? Journal of Language and Social
Psychology 21, 245–274.
Rockwell, P., 2000a. Actors, partners, and observers perceptions of
sarcasm. Perceptual and Motor Skills 91, 665–668.
Rockwell, P., 2000b. Lower, slower, louder: vocal cues of sarcasm. Journal
of Psycholinguistic Research 29, 483–495.
Rockwell, P., 2001. Facial expression and sarcasm. Perceptual and Motor
Skills 93, 47–50.
Rockwell, P., 2005. Sarcasm on television talk shows: determining speaker
intent through verbal and nonverbal cues. In: Clark, A. (Ed.),
Psychology of Moods. Nova Science Publishers Inc., New York, pp.
109–140.
Sabbagh, M.A., 1999. Communicative intentions and language: evidence
from right-hemisphere damage and autism. Brain Language 70, 29–69.
Schaffer, R., 1982. Are there consistent vocal clues for irony? In: Masek,
C.S., Hendrick, R.A., Miller, M.F. (Eds.), Parasession on Language
and Behavior. Chicago Linguistic Society, Chicago, IL, pp. 204–210.
Scherer, K.R., 1986. Vocal affect expression: a review and a model for
future research. Psychological Bulletien 99, 143–165.
Scherer, K.R., Ladd, D.R., Silverman, K.E., 1984. Vocal cues to speaker
affect: testing two models. Journal of the Acoustical Society of
America 76 (5), 1346–1356.
381
Sobin, C., Alpert, M., 1999. Emotion in speech: The acoustic attributes of
fear, anger, sadness, and joy. Journal of Psycholinguistic Research 28
(4), 1999.
Sperber, D., 1984. Verbal irony: Pretense or echoic mention? Journal of
Experimental Psychology: General 113 (1), 130.
Suls, J., 1972. A two-stage model for the appreciation of jokes and
cartoons: An information-processing analysis. In: Goldstein, J.H.,
McGhee, P.E. (Eds.), The Psychology of Humor. Academic Press,
New York, pp. 81–100.
Suls, J., 1983. Cognitive processes in humor appreciation. In: McGhee,
P.E., Goldstein, J.H. (Eds.), Handbook of Humor Research. Springer-
Verlag, New York, pp. 39–57.
Tartter, V.C., Braun, D., 1994. Hearing smiles and frowns in normal and
whisper registers. Journal of the Acoustical Society of America 96 (4),
2101–2107.
Tompkins, C.A., 1991. Automatic and effortful processing of emotional
intonation after right or left hemisphere brain damage. Journal of
Speech and Hearing Research 34 (4), 820–830.
Williams, C.L., Stevens, K.N., 1972. Emotions and speech: Some
acoustical correlates. Journal of the Acoustical Society of America
52 (4), 1238–1250.
Wilson, D., Wharton, T., 2006. Relevance and prosody. Journal of
Pragmatics 38, 1559–1579.
Winner, E., Leekman, S., 1991. Distinguishing irony from deception:
Understanding the speaker_s second-order intention. British Journal
of Developmental Psychology 9, 257–270.
Yoshida, H., Furuya, Y., Shimodaira, K., Kanazawa, T., Kataoka, R.,
Takahashi, K., 2000. Spectral characteristics of hypernasality in
maxillectomy patients. Journal of Oral Rehabilitation 27, 723–730.
Yumoto, E., Gould, W.J., Baer, T., 1982. Harmonics-to-noise ratio as an
index of the degree of hoarseness. Journal of the Acoustical Society of
America 71 (6), 1544–1550.